import cv2 import torch import random import numpy as np import PIL from PIL import Image from typing import Tuple import diffusers from diffusers.utils import load_image from diffusers.models import ControlNetModel from diffusers.pipelines.controlnet.multicontrolnet import MultiControlNetModel # from huggingface_hub import hf_hub_download from insightface.app import FaceAnalysis from style_template import styles from pipeline_stable_diffusion_xl_instantid_full import StableDiffusionXLInstantIDPipeline, draw_kps from controlnet_aux import OpenposeDetector import torch.nn.functional as F from torchvision.transforms import Compose import os # global variable device = "cuda" if torch.cuda.is_available() else "cpu" dtype = torch.float16 if str(device).__contains__("cuda") else torch.float32 STYLE_NAMES = list(styles.keys()) DEFAULT_STYLE_NAME = "Spring Festival" class EndpointHandler(): def __init__(self, model_dir): # hf_hub_download(repo_id="InstantX/InstantID", filename="ControlNetModel/config.json", local_dir="./checkpoints") # hf_hub_download( # repo_id="InstantX/InstantID", # filename="ControlNetModel/diffusion_pytorch_model.safetensors", # local_dir="./checkpoints", # ) # hf_hub_download(repo_id="InstantX/InstantID", filename="ip-adapter.bin", local_dir="./checkpoints") # Load face encoder # self.app = FaceAnalysis( # name="antelopev2", # root="./", # providers=["CPUExecutionProvider"], # ) dir_path = os.path.join("", "models", "antelopev2") print(dir_path) print(model_dir) self.app = FaceAnalysis(name='antelopev2', root='./', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']) self.app.prepare(ctx_id=0, det_size=(640, 640)) openpose = OpenposeDetector.from_pretrained("lllyasviel/ControlNet") # Path to InstantID models face_adapter = f"/repository/checkpoints/ip-adapter.bin" controlnet_path = f"/repository/checkpoints/ControlNetModel" # face_adapter = f"./checkpoints/ip-adapter.bin" # controlnet_path = f"./checkpoints/ControlNetModel" # Load pipeline face ControlNetModel self.controlnet_identitynet = ControlNetModel.from_pretrained( controlnet_path, torch_dtype=dtype ) # controlnet-pose controlnet_pose_model = "thibaud/controlnet-openpose-sdxl-1.0" controlnet_canny_model = "diffusers/controlnet-canny-sdxl-1.0" controlnet_pose = ControlNetModel.from_pretrained( controlnet_pose_model, torch_dtype=dtype ).to(device) controlnet_canny = ControlNetModel.from_pretrained( controlnet_canny_model, torch_dtype=dtype ).to(device) def get_canny_image(image, t1=100, t2=200): image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR) edges = cv2.Canny(image, t1, t2) return Image.fromarray(edges, "L") self.controlnet_map = { "pose": controlnet_pose, "canny": controlnet_canny } self.controlnet_map_fn = { "pose": openpose, "canny": get_canny_image } pretrained_model_name_or_path = "wangqixun/YamerMIX_v8" self.pipe = StableDiffusionXLInstantIDPipeline.from_pretrained( pretrained_model_name_or_path, controlnet=[self.controlnet_identitynet], torch_dtype=dtype, safety_checker=None, feature_extractor=None, ).to(device) self.pipe.scheduler = diffusers.EulerDiscreteScheduler.from_config( self.pipe.scheduler.config ) # load and disable LCM self.pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl") self.pipe.disable_lora() self.pipe.cuda() self.pipe.load_ip_adapter_instantid(face_adapter) self.pipe.image_proj_model.to("cuda") self.pipe.unet.to("cuda") def __call__(self, data): def convert_from_cv2_to_image(img: np.ndarray) -> Image: return Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB)) def convert_from_image_to_cv2(img: Image) -> np.ndarray: return cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR) def resize_img( input_image, max_side=1280, min_side=1024, size=None, pad_to_max_side=False, mode=PIL.Image.BILINEAR, base_pixel_number=64, ): w, h = input_image.size if size is not None: w_resize_new, h_resize_new = size else: ratio = min_side / min(h, w) w, h = round(ratio * w), round(ratio * h) ratio = max_side / max(h, w) input_image = input_image.resize([round(ratio * w), round(ratio * h)], mode) w_resize_new = (round(ratio * w) // base_pixel_number) * base_pixel_number h_resize_new = (round(ratio * h) // base_pixel_number) * base_pixel_number input_image = input_image.resize([w_resize_new, h_resize_new], mode) if pad_to_max_side: res = np.ones([max_side, max_side, 3], dtype=np.uint8) * 255 offset_x = (max_side - w_resize_new) // 2 offset_y = (max_side - h_resize_new) // 2 res[ offset_y : offset_y + h_resize_new, offset_x : offset_x + w_resize_new ] = np.array(input_image) input_image = Image.fromarray(res) return input_image def apply_style( style_name: str, positive: str, negative: str = "" ) -> Tuple[str, str]: p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME]) return p.replace("{prompt}", positive), n + " " + negative face_image_path = data.pop("face_image_path", "https://i.ibb.co/GQzm527/examples-musk-resize.jpg") pose_image_path = data.pop("pose_image_path", "https://i.ibb.co/TRCK4MS/examples-poses-pose2.jpg") style_name = data.pop("style_name", DEFAULT_STYLE_NAME) prompt = data.pop("inputs", "a man flying in the sky in Mars") negative_prompt = data.pop("negative_prompt", "(lowres, low quality, worst quality:1.2), (text:1.2), watermark, (frame:1.2), deformed, ugly, deformed eyes, blur, out of focus, blurry, deformed cat, deformed, photo, anthropomorphic cat, monochrome, pet collar, gun, weapon, blue, 3d, drones, drone, buildings in background, green") identitynet_strength_ratio = 0.8 adapter_strength_ratio = 0.8 pose_strength = 0.5 canny_strength = 0.3 num_steps = 20 guidance_scale = 5.0 controlnet_selection = ["pose", "canny"] scheduler = "EulerDiscreteScheduler" self.pipe.disable_lora() scheduler_class_name = scheduler.split("-")[0] add_kwargs = {} if len(scheduler.split("-")) > 1: add_kwargs["use_karras_sigmas"] = True if len(scheduler.split("-")) > 2: add_kwargs["algorithm_type"] = "sde-dpmsolver++" scheduler = getattr(diffusers, scheduler_class_name) self.pipe.scheduler = scheduler.from_config(self.pipe.scheduler.config, **add_kwargs) # apply the style template prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt) face_image = load_image(face_image_path) face_image = resize_img(face_image, max_side=1024) face_image_cv2 = convert_from_image_to_cv2(face_image) height, width, _ = face_image_cv2.shape # Extract face features face_info = self.app.get(face_image_cv2) # if len(face_info) == 0: # raise gr.Error( # f"Unable to detect a face in the image. Please upload a different photo with a clear face." # ) face_info = sorted( face_info, key=lambda x: (x["bbox"][2] - x["bbox"][0]) * x["bbox"][3] - x["bbox"][1], )[ -1 ] # only use the maximum face face_emb = face_info["embedding"] face_kps = draw_kps(convert_from_cv2_to_image(face_image_cv2), face_info["kps"]) img_controlnet = face_image if pose_image_path is not None: pose_image = load_image(pose_image_path) pose_image = resize_img(pose_image, max_side=1024) img_controlnet = pose_image pose_image_cv2 = convert_from_image_to_cv2(pose_image) face_info = self.app.get(pose_image_cv2) # if len(face_info) == 0: # raise gr.Error( # f"Cannot find any face in the reference image! Please upload another person image" # ) face_info = face_info[-1] face_kps = draw_kps(pose_image, face_info["kps"]) width, height = face_kps.size control_mask = np.zeros([height, width, 3]) x1, y1, x2, y2 = face_info["bbox"] x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2) control_mask[y1:y2, x1:x2] = 255 control_mask = Image.fromarray(control_mask.astype(np.uint8)) controlnet_scales = { "pose": pose_strength, "canny": canny_strength } self.pipe.controlnet = MultiControlNetModel( [self.controlnet_identitynet] + [self.controlnet_map[s] for s in controlnet_selection] ) control_scales = [float(identitynet_strength_ratio)] + [ controlnet_scales[s] for s in controlnet_selection ] control_images = [face_kps] + [ self.controlnet_map_fn[s](img_controlnet).resize((width, height)) for s in controlnet_selection ] generator = torch.Generator(device=device).manual_seed(42) print("Start inference...") print(f"[Debug] Prompt: {prompt}, \n[Debug] Neg Prompt: {negative_prompt}") self.pipe.set_ip_adapter_scale(adapter_strength_ratio) images = self.pipe( prompt=prompt, negative_prompt=negative_prompt, image_embeds=face_emb, image=control_images, control_mask=control_mask, controlnet_conditioning_scale=control_scales, num_inference_steps=num_steps, guidance_scale=guidance_scale, height=height, width=width, generator=generator, ).images return images[0]